The image is making a promise before anyone reads the caption
You ask an image tool for a cleaner background around a bottle. It returns better lighting, a richer surface and a premium-looking label. It also changes the cap, removes a line of text and makes the bottle look twice as large beside the hand holding it.
The image is attractive. The product is wrong. This is a different problem from deciding whether AI imagery looks tasteful. Customers use pictures to infer shape, quantity, scale, finish and what arrives in the package.
Start the review by naming what must remain true. For a real product, the approved reference may be a current product photograph, packaging file and specification sheet. A generated concept can have more freedom, but it should be presented as a concept. A small disclosure does not repair a misleading product representation.
Build a truth sheet before generating variations
Take one approved product and list the features that affect a buyer's expectation: dimensions, material, colour variant, count, label, closure, included accessories and packaging. Mark which features the proposed image actually shows. If a dimension is not visible, the visual may still imply it through a nearby object.
Separate fixed facts from creative choices. Fixed facts include the product's shape and the number of items in the pack. Creative choices may include background colour, lighting direction and empty space for a headline. Some changes sit between them: a dramatic reflection can make matte plastic look like polished metal.
Keep the reference version with the task. A discontinued label or earlier packaging photograph is not an acceptable truth source for the current listing. If the product team cannot identify the current reference, resolve that first. More precise image prompts cannot fix disagreement about what is being sold.
A fictional image review with three different outcomes
Imagine a fictional 500 ml refillable bottle sold individually with one screw cap. The approved source shows an uncoated metal finish and a printed care symbol. A generated lifestyle image shows two bottles, a carry sleeve and a glossy finish.
| Finding | Why it matters | Decision |
|---|---|---|
| Second bottle prominent | Could imply a two-pack | Remove it or clearly establish a composition that does not misstate quantity |
| Carry sleeve invented | Looks like an included accessory | Remove; it is not sold with this product |
| Finish looks glossy | Changes expected material appearance | Restore an accurate finish from approved reference |
| Background has better light | Does not change the product fact | Keep if the final appearance remains accurate |
This is a teaching scenario, not a legal finding or a DLVX campaign result. The point is to distinguish a creative improvement from a changed product claim. “Looks more premium” is not a sufficient reason to accept either.
Review in passes, not one general glance
Identity pass: compare silhouette, proportions, cap, seams, connectors and visible branding. Zoom in. If the tool invented small text, replace it using the approved artwork or use the real source asset. Do not ask a reviewer to overlook a distorted label because most visitors will see a small thumbnail.
Quantity and scale pass: count the units, identify accessories and examine reference objects. A hand, chair or countertop can imply size even without a measurement label. Check whether the crop makes a small item resemble a larger model.
Material and use pass: inspect texture, reflection and the action shown. The image should not imply a capability or use the product does not support. A visually persuasive scene is especially dangerous when it looks like performance evidence.
Context pass: read the headline, price, variant selector and image together. A truthful photograph of one variant can still mislead if it appears beside another variant's description.
Write a correction the editor can actually apply
“Make it more realistic” gives the editor no acceptance standard. A better note says: “Keep the background and camera angle. Restore the bottle silhouette, cap and printed care symbol from reference A. Show one bottle only. Remove the sleeve. Match the matte finish. Keep sufficient space for the existing headline.”
After revision, compare the whole image again. Fixing the cap may change the label or create a new reflection. Do not limit the second review to the defect mentioned in the feedback. Keep a short revision log so the next reviewer knows which product facts required protection.
If repeated generation keeps changing the product, stop using generation for that part. Use the approved product photograph with controlled background editing, conventional compositing or a new photograph. The goal is a publishable truthful asset, not proving that one tool can eventually obey every constraint.
Motion adds another place for the product to change
A still may pass while an animated version morphs the logo, changes the number of parts or reveals an invented reverse side. Review the beginning, middle and end, plus transitions where the product turns or an object enters the frame. A short loop can hide a defect if nobody watches it carefully.
Use motion when it teaches something: how a component opens, how two parts connect, or how a real workflow progresses. If the motion only adds movement around a static product, compare it with the approved still. The still may communicate faster and cost less to review.
Keep a static equivalent for the same placement. Make sure motion does not obscure the label, price or disclosure. Test the actual mobile crop rather than assuming the desktop composition survives a vertical format.
Keep the publishing decision with the asset
Product and variant: Approved reference files and version: Fixed facts that must remain visible and true: Permitted creative changes: Identity check: Quantity, accessories and scale check: Material and demonstrated-use check: Final crop, headline and variant match: Source rights and required disclosure review: Accepted file, placement, reviewer and date: Rejected versions and reason:
Rights, platform rules and disclosure obligations depend on the asset and where it will be used; this checklist does not determine legal compliance. It addresses the product-truth review that can be performed before those publication checks. NIST's framework provides broader context for identifying and managing AI-related risks, not approval of any image.
Choose one generated image currently under consideration. Place it beside the current product reference and write three things a buyer could infer from it. Verify each inference before approving the crop.
Sources and dates
Sources checked on 20 October 2026. Worked examples are teaching examples unless explicitly attributed. Provider documentation describes product behaviour; the operating recommendations are DLVX editorial judgement.
- NIST: AI Risk Management FrameworkA voluntary framework for managing AI risks; it does not certify the processes or examples in this guide. Checked 2026-10-20.